Research On Algorithm Of Counting High-density Crowd

The world population is growing very quickly in recent years. In the public places,unfortunate accidents caused by high-density crowd occur frequently. At the same time, thevideo surveillance systems are ubiquitous. If we make use of the existing resources, theseintelligent systems can effectively forewarn and avoid disaster events. Compared with thetraditional approach, the intelligent system of counting and density estimation can alsoimprove the utilization rate of public facilities, and arrange the allocation of manpower andmaterial resources.The main content of this paper is counting high-density crowd. First of all, theforeground image of moving target is segmented from the image. And then, extracting featureparameters of foreground image. Finally, the crowd’s number and density is acquired byregression.To get the foreground image, this paper, firstly, gray and smooth the input image. Thengetting foreground image by background subtraction operation and moving average method.And also morphology processing was performed on the binary image to eliminate noise.In feature extraction, feature vector is made up with the area foreground pixel, the fourimportant Gray Level Co-occurrence Matrix characteristics (energy, entropy, contrast,homogeneity) and the number of SURF feature points. Linear interpolation weightsperspective correction method is considered for camera deformity correction.The feature vector after correction is trained by Support Vector Regression. SVR modelis established to estimate the crowd number.Finally, this paper studied on two different videos to verify the effectiveness of thealgorithm. Experimental results show that, this method has strong adaptability and improvesthe accuracy of detection. The average error is less than2people per frame. It can effectivelyestimate the density and the number of crowd.